In a groundbreaking event at Fort Liberty in 2020, an artificial intelligence program identified and suggested targets based on satellite images. After human confirmation, the AI instructed an M142 Himars rocket launcher to fire at a decommissioned tank, marking the first time American soldiers had struck a target located and identified by an AI program. Colonel Joseph O’Callaghan, the fire-support coordinator for the 18th Airborne Corps, described the event as demonstrating “the art of the possible” and highlighted the significant role AI could play in military operations.
In less than four years since the AI milestone at Fort Liberty, the United States has moved from theoretical use to practical application of AI in warfare. The computer vision algorithms of the Department of Defense’s Project Maven have recently located rocket launchers in Yemen, identified surface vessels in the Red Sea, and assisted in targeting decisions for strikes in Iraq and Syria. Other countries, including Israel and Ukraine, are also leveraging AI in military operations, marking a significant advancement in the integration of artificial intelligence technologies on the battlefield.
The transition of AI from the laboratory to the battlefield poses complex challenges for military leaders. Advocates emphasize the need for rapid adoption, anticipating combat scenarios surpassing the speed of human decision-making. However, concerns exist regarding the readiness of military networks and data, the reluctance of frontline troops to rely on untested software, and ethical considerations surrounding the delegation of potentially life-altering decisions to machines. The debate reflects the tension between technological advancements, operational preparedness, and ethical considerations, amid geopolitical competition, especially with China’s AI ambitions.
The 18th Airborne Corps serves as a major test bed for Project Maven, utilizing powerful AI algorithms to identify personnel and equipment on the battlefield. Leveraging machine learning breakthroughs, the system autonomously learns to recognize objects based on training data and user feedback. AI models can discern significant changes to objects, indicating potential developments, and integrate this information with satellite imagery and geolocation data in the Maven Smart System. The Pentagon’s substantial budget allocation for AI-related activities highlights its recognition of AI’s transformative potential in modern warfare, raising ethical and strategic considerations amid global AI military integration.
AI has a historical presence in the Defense Department, with early implementations like the Semi-Automatic Ground Environment air-defense system during the Cold War. Operation Desert Storm saw the use of analysis tools for troop movement planning. While AI advanced in the civilian sector, the military faced challenges in understanding machine learning, lacking mass cloud storage, and computer-friendly data by 2016. The Pentagon recognized the urgency and allocated substantial funds for AI development, leading to projects like Maven, transforming warfare capabilities and posing ethical and strategic implications.
Concerns about the military’s slow development of tools led to Project Maven, initially proposed by Will Roper to apply machine learning to automatic target recognition with a $50 million budget. Maven aimed to assess object recognition tools from various vendors, but initial testing revealed challenges, including hazy images, inconsistent labeling, and difficulties in handling real-world field conditions. Google’s involvement faced internal protests, and the Pentagon later exempted Maven from public disclosure to protect sensitive information. These challenges highlighted the complexities of integrating AI into military operations.
Despite initial skepticism among soldiers, various technology and defense companies, including Palantir Technologies, Amazon Web Services, ECS Federal, L3Harris Technologies, Maxar Technologies, Microsoft, and Sierra Nevada, continued their involvement with Project Maven. Soldiers, like Joey Temple, a senior targeting officer, initially doubted the system’s effectiveness but changed their minds during real-world scenarios. Temple, who used Maven Smart System during the evacuation from Afghanistan, appreciated its ability to consolidate data feeds and provide valuable insights during complex operations.
Project Maven has evolved significantly, incorporating data from various sources, including radar systems, infrared sensors, and non-visual information from electronic surveillance and social media feeds. The platform’s advancements allow it to analyze complex scenarios, identify objects of interest, and contribute to decision-making processes. Soldiers like Joey Temple acknowledge the platform’s efficiency, with the ability to process and sign off on targets more quickly with Maven’s assistance. Despite the technology’s support, there is still a reliance on human decision-making to ensure accuracy and avoid potential errors.
After Russia invaded Ukraine in February 2022, the 18th’s AI engineering efforts took on new urgency. O’Callaghan, Temple and 270 others from the corps headquarters had deployed to a garrison in Germany, taking over a former racquetball hall as a command center. Maven Smart System was open on many of their screens. An information operations and civil affairs officer, who asked not to be named because she often deploys on special operations missions, became one of those using the software to brief Army commanders.
The officer’s focus was on judging whether Ukrainians in particular areas had the will to resist Russian forces. “When you layer things, different things begin to jump out at you,” she says. “I would have pins dropped on every event that was resistance-related on the map.” These ranged from the mundane—sightings of blue and yellow ribbons tied to fence posts—to the violent, such as attacks on Russian officials in occupied zones. These details, along with many others, then made their way into the classified system US commanders use to inform their understanding of events on the ground.
O’Callaghan and Temple declined to detail how the Pentagon is employing AI systems such as Maven to support Ukraine. But people familiar with the operations, who asked not to be named given the subject’s sensitivity, say the US has used satellite intelligence and Maven Smart System to supply the locations of Russian equipment to Ukrainian forces, who have then targeted those assets with GPS-guided missiles. One of the people says that aiding Kyiv has also helped the Pentagon get much better at using AI tools—and more confident in its forecasts of how they might be used in a conflict with China. (During the first 10 months supporting Ukraine, Maven underwent more than 50 rounds of improvement, according to another person with knowledge of its development.)
I was given a demonstration of Maven when I visited the 18th last year. It was drawn from a real-life operation in early 2023, when a US Navy ship evacuated American citizens from Sudan, which was in the midst of a civil war. The system’s central map was generated, in the case of my demonstration, from unclassified satellite imagery. On the surface, yellow boundary boxes marked where algorithms had identified ships in the Gulf of Aden. Blue areas signified places that would be included on a no-strike list, such as hospitals and schools. On the left side of the screen, icons offered separate data streams, such as vessel tracking feeds, that could be overlaid on the map. There was also a function for a “tactical data link”: a method for transmitting a commander’s decision to fire a weapon directly between machines.